CN116738072B - Multidimensional recommendation method combining human factor information - Google Patents
Multidimensional recommendation method combining human factor information Download PDFInfo
- Publication number
- CN116738072B CN116738072B CN202311026200.XA CN202311026200A CN116738072B CN 116738072 B CN116738072 B CN 116738072B CN 202311026200 A CN202311026200 A CN 202311026200A CN 116738072 B CN116738072 B CN 116738072B
- Authority
- CN
- China
- Prior art keywords
- user
- users
- attention
- recommendation
- similarity
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 33
- 230000004424 eye movement Effects 0.000 claims abstract description 37
- 238000011156 evaluation Methods 0.000 claims abstract description 32
- 238000013528 artificial neural network Methods 0.000 claims abstract description 10
- 230000004927 fusion Effects 0.000 claims description 16
- 238000005457 optimization Methods 0.000 claims description 13
- 230000003993 interaction Effects 0.000 claims description 12
- 238000004364 calculation method Methods 0.000 claims description 10
- 239000002245 particle Substances 0.000 claims description 8
- 230000008569 process Effects 0.000 claims description 8
- 238000013507 mapping Methods 0.000 claims description 6
- 238000012545 processing Methods 0.000 claims description 6
- 230000006399 behavior Effects 0.000 claims description 4
- 238000001914 filtration Methods 0.000 claims description 3
- 238000010606 normalization Methods 0.000 claims description 3
- 210000001747 pupil Anatomy 0.000 claims description 3
- 238000012549 training Methods 0.000 claims description 3
- KJONHKAYOJNZEC-UHFFFAOYSA-N nitrazepam Chemical compound C12=CC([N+](=O)[O-])=CC=C2NC(=O)CN=C1C1=CC=CC=C1 KJONHKAYOJNZEC-UHFFFAOYSA-N 0.000 claims description 2
- 238000000611 regression analysis Methods 0.000 claims description 2
- 238000012163 sequencing technique Methods 0.000 claims 1
- 208000003443 Unconsciousness Diseases 0.000 abstract description 2
- 230000006870 function Effects 0.000 description 10
- 238000010586 diagram Methods 0.000 description 9
- 238000002474 experimental method Methods 0.000 description 8
- 230000000694 effects Effects 0.000 description 5
- 238000005516 engineering process Methods 0.000 description 5
- 230000008713 feedback mechanism Effects 0.000 description 4
- 230000002776 aggregation Effects 0.000 description 3
- 238000004220 aggregation Methods 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 3
- 230000019771 cognition Effects 0.000 description 3
- 230000008451 emotion Effects 0.000 description 3
- 230000003542 behavioural effect Effects 0.000 description 2
- 210000004556 brain Anatomy 0.000 description 2
- 230000005611 electricity Effects 0.000 description 2
- 230000003749 cleanliness Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 235000003642 hunger Nutrition 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000003064 k means clustering Methods 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000037351 starvation Effects 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9536—Search customisation based on social or collaborative filtering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/27—Regression, e.g. linear or logistic regression
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/004—Artificial life, i.e. computing arrangements simulating life
- G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Databases & Information Systems (AREA)
- Biomedical Technology (AREA)
- Health & Medical Sciences (AREA)
- Probability & Statistics with Applications (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses a multidimensional recommendation method combining human factor information, and relates to the technical field of computer science. Comprising the following steps: determining neighbor users based on the multi-dimensional similarity fused by the preference weights of the users; constructing a user attention model based on the eye movement signal, and acquiring the attention of a user when browsing the project; based on the depth neural network, merging the multidimensional scores and the attention degree, and generating a recommendation result; and constructing a user evaluation feedback model based on the eye movement signals, acquiring the evaluation of the user on the recommendation result and optimizing the recommendation algorithm. According to the invention, the human factor engineering is introduced into a recommendation algorithm, the unconscious behavior and physiological signals and other human factor information when a user browses the project are analyzed, a user attention model is constructed, the attention of the user to the project is obtained, multidimensional scoring information and attention information are fused, the user preference and attention are analyzed, and the accuracy of a recommendation result is improved. In addition, an evaluation feedback model is constructed by using the human factor information, the evaluation of the user on the recommendation result is obtained, the user preference is corrected, and the recommendation algorithm is optimized.
Description
Technical Field
The invention relates to the technical field of computer science, in particular to a multidimensional recommendation method combining human factor information.
Background
With the development of information technology and the Internet, people gradually walk into the information overload age from the information starvation age. In order to solve the problem of information overload, a recommendation system has been attracting attention in recent years. The core technology of the recommendation system is a recommendation algorithm, and in order to conduct fine granularity analysis on diversity of user preference, the multidimensional recommendation algorithm becomes a research hotspot.
The current interaction information between the user and the item mainly considers explicit information (such as user scores and the like) or implicit information (browsing duration, comments and the like). However, in an actual recommendation scene, there are often problems that the user needs are unknown, the evaluation is fuzzy, the preference is difficult to capture, and the like, for example, some users have fuzzy self needs, the types of the contacted items are not comprehensive enough, and subjective scoring information cannot accurately reflect the favorite degree of the user on the rated items; in addition, minors or some special groups have limited expression capability, and their own preference is not clear enough to give accurate evaluation. On the other hand, the recommendation algorithm often adopts indexes such as accuracy and recall rate to perform performance evaluation, the indexes are obtained through offline calculation, the actual satisfaction degree of a user on a recommendation result is difficult to reflect, personalized evaluation of the user on a recommendation system cannot be obtained, effective feedback information cannot be formed, and the recommendation algorithm is improved.
Therefore, how to provide a method for more comprehensively considering the interaction information of the user, accurately reflecting the preference of the user, improving the recommendation accuracy, and obtaining the personalized evaluation and feedback of the user is a problem to be solved in the art. In addition, the user has a emphasis on different attributes of the item, and how to consider the diversity of user preference in the recommendation algorithm, so that the improvement of the algorithm efficiency is also a key problem.
Disclosure of Invention
In view of the above, the present invention provides a multidimensional recommendation method combining human factor information, which solves the problems existing in the background technology.
In order to achieve the above object, the present invention provides the following technical solutions:
a multidimensional recommendation method combining human factor information comprises the following steps:
determining neighbor users based on the multi-dimensional similarity fused by the preference weights of the users;
constructing a user attention model based on the eye movement signal, and acquiring the attention of a user when browsing the project;
based on the depth neural network, merging the multidimensional scores and the attention degree, and generating a recommendation result;
and constructing a user evaluation feedback model based on the eye movement signals, acquiring the evaluation of the user on the recommendation result, and forming feedback information to correct the user preference.
Optionally, determining the neighboring user specifically includes the following steps:
based on the multidimensional scoring information of the users, carrying out user classification by adopting a clustering algorithm;
and calculating the similarity of each dimension of various users, carrying out multi-dimension similarity weight optimization of the users by adopting a particle swarm optimization algorithm, and obtaining the total similarity of the users by weighting and fusion to obtain a neighbor user set.
Optionally, the specific process of user classification is as follows:
based on the Euclidean distance between the multidimensional score and the total score of the user, obtaining a preference value of each user:
（1）；
wherein:ufor the user to be the target user,scoring the population of items for the target user, +.>Item for target userkScoring of the individual dimensions; />For usersuA set of scored items; />For usersuFor the firstkPreference values for the individual dimensions;the Euclidean distance between the multidimensional scores and the total scores of the users is calculated;
calculating preference values of all users for each dimension, and carrying out normalization processing to obtain preference characteristics of the users;
based on preference characteristics of users, the users are clustered by adopting a K-means algorithm.
Optionally, the specific process of acquiring the neighbor user set is as follows:
and calculating the similarity of each dimension between users in the same category by adopting the Pelson coefficient:
（2）；
wherein:representing a useruAnd a uservIn dimension ofcSimilarity of (2); />And->Respectively represent usersuAnd a uservFor projectsiIn dimension ofcScoring; />Representing a useruAnd a uservA set of commonly scored items;and->Respectively represent usersuAnd a uservIn item collections with respect to dimensionscIs a mean score of (2);
and carrying out similarity fusion according to the preference characteristics of the user, wherein the formula of the similarity fusion of the user is as follows:
（3）；
wherein:for usersuAnd a uservOverall similarity of->And->Respectively represent usersuAnd a uservIs characterized by comprising the following steps of (1) each dimension similarity and corresponding weight;
optimizing the similarity fusion weight of each dimension by adopting a particle swarm optimization algorithm, taking the minimum value of the average absolute error between the predicted score and the actual score of the user as an optimized objective function, wherein the fitness function is as follows:
（4）；
wherein:and->Respectively represent usersuAnd a uservFor projectsiIs used to determine the actual score and the predicted score of (c),Nto predict the number of items.
Optionally, the method for acquiring the attention degree when the user browses the item is as follows: analyzing interaction behaviors and eye movement signals when a user browses items, and establishing a nonlinear relation between the eye movement signals and the attention of the user; the method specifically comprises the following steps:
assume thatFor inputting sample data, wherein->Representing the user's attention, ->As eye movement data samples, four eye movement index features including gazing time length, gazing point number, pupil diameter and eye jump number are adopted as sample data through correlation regression analysis;
the expression of the regression function is:
（5）；
wherein:is a nonlinear mapping function and is used for mapping training data to a high-dimensional linear space; />Andbis a model parameter;
introducing relaxation variablesAnd->Converting the nonlinear support vector regression problem into a quadratic programming problem:
（6）；
wherein:，/>for the number of samples,Cfor punishment factors->Is the insensitivity loss coefficient;
and introducing Lagrange multipliers, solving parameters in the quadratic programming problem, and further obtaining the attention degree of the user to the project.
Optionally, generating a recommendation result specifically includes the following steps:
processing and fusing the user multidimensional scoring information and the user attention in each category by using a deep neural network to obtain the comprehensive score of the target user on the project;
in the neighbor user set, predicting the comprehensive score of the target user on the non-contact item by a collaborative filtering algorithm, wherein the calculation formula is as follows:
（7）；
wherein:representing a target useruFor articlesiIs a predictive score of (2); />Representing a target useruIs a comprehensive score average value of (2);Nrepresenting a set of neighbor users of the target user; />For usersuAnd a uservOverall similarity of (c); />And->Respectively represent neighboring usersvFor articlesiIs a comprehensive score and user of (2)vIs a comprehensive score average value of (2);
and sorting the items based on the predicted comprehensive scores of the items by the users, and selecting top-N items as recommendation results formed by the users.
Optionally, the input layer of the deep neural network is a user multidimensional score and a degree of attention, the hidden layer is a 2-layer, and the output of the output layer is a user comprehensive score.
Optionally, the method includes the steps of obtaining the user's evaluation of the recommendation result, forming feedback information to correct the user's preference, and specifically including the following steps:
after a user obtains a recommendation result, a user evaluation feedback model is constructed based on the eye movement signals, the eye movement signals when the user browses the recommendation result are tracked and analyzed in real time, the satisfaction degree of the user on the recommendation result is obtained, and then personalized evaluation of the user is obtained;
and taking the evaluation result as feedback information, updating interaction information between the user and the project, correcting user preference, and optimizing a recommendation algorithm.
Compared with the prior art, the multi-dimensional recommendation method combining the human factor information is provided, on one hand, the behavioral information and the physiological signal of the user are captured through the human factor engineering technology, and the psychological activities, conscious emotion and attention degree of the user when browsing the project are perceived, so that a more comprehensive, objective and accurate user portrait is established; on the other hand, the invention combines the personal information to acquire the personalized evaluation of the user on the recommendation result, establishes a feedback mechanism to update the interaction information of the user and the project, corrects the user preference, further optimizes the recommendation algorithm, and is beneficial to forming the recommendation result meeting the real requirement of the user. In addition, the method and the device cluster the users according to the preference, so that the calculation efficiency is improved, the preference weights of the users are fused by using an intelligent algorithm, and the accuracy of the recommendation result is improved.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings that are required to be used in the embodiments or the description of the prior art will be briefly described below, and it is obvious that the drawings in the following description are only embodiments of the present invention, and that other drawings can be obtained according to the provided drawings without inventive effort for a person skilled in the art.
FIG. 1 is a flow chart of a multi-dimensional recommendation method combining human factor information provided by the invention;
FIG. 2 is a flowchart of a clustering algorithm provided by the present invention;
fig. 3 is a schematic diagram of acquiring a neighboring user according to the present invention;
FIG. 4 is a functional block diagram of a user attention model constructed based on eye movement signals provided by the present invention;
FIG. 5 is a schematic diagram of a DNN-based user scoring and attention aggregation model provided by the present invention;
FIG. 6 is a graph showing comparison of average contour coefficients of clusters provided by the invention;
fig. 7 is a PSO iterative optimization diagram provided by the present invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
Because the unconscious behaviors of the user, physiological signals and other human factor information can accurately and objectively express the emotion and the attention of the person, and the cognitive process of the person is reflected. The ergonomic techniques may capture these behavioral information and physiological signals, fitting them to the user's attention. Therefore, aiming at the traditional multidimensional recommendation algorithm, only the implicit information of the user is considered, the user demand is unknown, the evaluation is fuzzy, the real preference is difficult to accurately capture, and the user personalized evaluation and feedback mechanism is lacked, so that the recommendation result is different from the user demand.
As shown in fig. 1, the basic idea of the multidimensional recommendation method combined with the human factor information disclosed in the embodiment of the present invention can be summarized as the following parts:
(1) Determining neighbor users based on the multi-dimensional similarity fused by the preference weights of the users;
based on the multidimensional scoring information of the users, carrying out user classification by adopting a clustering algorithm; and calculating the similarity of each dimension of various users, carrying out multi-dimension similarity weight optimization of the users by adopting a particle swarm optimization algorithm (Particle Swarm Optimization, PSO), and obtaining the total similarity of the users by weighting and fusion to obtain a neighbor user set.
Specifically, first, a preference value of each user is obtained based on the Euclidean distance between the multidimensional score and the total score of the user:
（1）；
wherein:ufor the user to be the target user,scoring the population of items for the target user, +.>Item for target userkScoring of the individual dimensions; />For usersuA set of scored items; />For usersuFor the firstkPreference values for the individual dimensions;the Euclidean distance between the multidimensional scores and the total scores of the users is calculated;
calculating preference values of all users for each dimension, and carrying out normalization processing to obtain preference characteristics of the users;
based on preference characteristics of users, the users are clustered by adopting a K-means algorithm. If the recommendation is performed according to the preference of each user, the calculation scale is too large, the recommendation efficiency is affected, and the users are clustered through the user multidimensional preference clustering schematic diagram shown in fig. 2, so that the subsequent recommendation task can be completed in small batches of users, the target users can obtain recommendation results which better accord with the preference of the target users, and the high efficiency of the recommendation algorithm is improved.
After classifying users according to multidimensional preferences, the user preferences of the same cluster class can be considered to be close. Referring to the schematic block diagram of the neighboring user obtaining a certain user category shown in fig. 3, under a certain user category, the similarity of the user in each dimension needs to be calculated, and the total similarity is calculated by combining with the preference weight, so as to obtain the neighboring user.
Since the pearson coefficients are insensitive to absolute values, the pearson coefficients can be well applied to the data sparseness condition. Thus, the pearson coefficients are used to calculate the degree of similarity in each dimension between users in the same category:
（2）；
wherein:representing a useruAnd a uservIn dimension ofcSimilarity of (2); />And->Respectively represent usersuAnd a uservFor projectsiIn dimension ofcScoring; />Representing a useruAnd a uservA set of commonly scored items;and->Respectively represent usersuAnd a uservIn item collections with respect to dimensionscIs a mean score of (2);
under the same category, the preferences of the users have similarity, and after the similarity of the users about each dimension under the same category is obtained, similarity fusion is needed according to the preference weight of the users, so that the similarity between the users can be more accurately obtained, and the formula of the similarity fusion of the users is as follows:
（3）；
wherein:for usersuAnd a uservOverall similarity of->And->Respectively represent usersuAnd a uservIs a function of the degree of similarity of the dimensions of the system and the corresponding preference weights;
optimizing the similarity fusion weight of each dimension by adopting a particle swarm optimization algorithm (Particle Swarm Optimization, PSO), taking the minimum value of the average absolute error (Mean Absolute Error, MAE) between the predicted score and the actual score of the user as a PSO optimizing target function, wherein the fitness function is as follows:
（4）；
wherein:and->Respectively represent usersuAnd a uservFor projectsiIs used to determine the actual score and the predicted score of (c),Nto predict the number of items.
From the above, when the PSO is utilized to optimize the preference weight of the user, multidimensional similarity fusion calculation can be performed. And under each user category, the accuracy of the overall similarity of the users is improved, the recall effect is finally optimized, and the scoring prediction accuracy is improved.
(2) Constructing a user attention model based on the eye movement signal, and acquiring the attention of a user when browsing the project;
for recommendation algorithms, analyzing the user's attention to the item is a key issue. At present, most recommendation algorithms only consider explicit data (scoring) and implicit data (clicking times, browsing time length) to conduct recommendation, and the information is difficult to accurately reflect real preferences of users, and finally recommendation results are affected. However, when the user browses the corresponding items using the recommendation system, corresponding physiological signals such as eye movement, brain electricity, electrocardiosignals and the like are generated, and the signals can objectively reflect the psychological activities and cognitive processes of the user. In addition, the human eye movement process can reflect the cognition degree and emotion attitude of the user when browsing the project. The eye movement tracking technology can carry out refined tracking on visual signals of the human body about external information, and is beneficial to accurately analyzing the attention of a user when browsing items.
Therefore, the interaction behavior and eye movement signals when the user browses the project are analyzed by using the human factor information of the user, the nonlinear relation between the eye movement signals and the attention of the user is established, and the attention of the user to the project can be objectively and accurately analyzed. Because the support vector regression (Support Vector Regression, SVR) model can fit the relationship between the dependent variable and the independent variable under the condition that the variable relationship is uncertain, and output prediction is realized, a nonlinear SVR model can be constructed based on the eye movement signals, the attention of the user to the project is predicted, and fig. 4 is a schematic block diagram of the construction of the user attention model based on the eye movement signals.
Specifically, assume thatFor inputting sample data, wherein->Representing the degree of attention of the user,for an eye movement data sample, adopting fixation time length and fixation point through correlation regression analysisFour eye movement index features of number, pupil diameter and eye jump number are used as sample data;
the expression of the regression function is:
（5）；
wherein:is a nonlinear mapping function and is used for mapping training data to a high-dimensional linear space; />Andbis a model parameter;
to enhance the robustness of the model, letMore approximate to the predicted value->Introducing a relaxation variable->And->Converting the nonlinear support vector regression problem into a quadratic programming problem:
（6）；
wherein:，/>for the number of samples,Cfor punishment factors->Is the insensitivity loss coefficient;
and introducing Lagrange multipliers, solving parameters in the quadratic programming problem, and further obtaining the attention degree of the user to the project.
In this embodiment, the SVR model may fit a nonlinear relationship between the eye movement parameter of the user and the attention, and may obtain the attention of the user to the item. Compared with uncertainty and ambiguity of explicit information such as scoring, the user attention regression model based on human-computer interaction and eye tracking can better reflect the interests of the user, and the accuracy and the interpretability of the algorithm can be improved by considering the attention of the user to the item in the recommendation algorithm.
(3) Based on the depth neural network, merging the multidimensional scores and the attention degree, and generating a recommendation result;
in order to fully analyze the scoring information and eye movement information of the user and improve the scoring prediction accuracy, the multi-dimensional scoring of the user on the item and the attention degree of the user on the item are required to be fused in each user category, and the scoring of the target user on the non-contact item is predicted in the recalled items of the neighbor user set. The Deep neural network (Deep-Learning Neural Network, DNN) can extract and process the interaction information characteristics of the user and the project in a large amount of data, and the DNN is adopted to fuse the scoring information of the project and the attention information of the user, so that the nonlinear relation between the scoring information of the user and the comprehensive scoring information of the user can be fitted well. Fig. 5 is a schematic diagram of a DNN-based user multidimensional scoring and attention aggregation model.
Specifically, in fig. 5, DNN needs to process and aggregate the user multidimensional scoring information and the user attention information in each category to obtain a comprehensive score of the target user for the item. The multidimensional scoring and the attention of the user to the project are input into DNN after data processing; for fitting the nonlinear relation between the multidimensional scoring information of the user and the attention and the comprehensive scoring, setting a DNN hidden layer as 2 layers; and the output of the output layer is the comprehensive score of the user as the multi-dimensional score information and the interest degree of each user are fused and the comprehensive score of the user is obtained.
After obtaining the comprehensive score of the user on the project, predicting the comprehensive score of the target user on the non-contact project in the neighbor user set by a collaborative filtering algorithm, wherein the calculation formula is as follows:
（7）；
wherein:representing a target useruFor articlesiIs a predictive score of (2); />Representing a target useruIs a comprehensive score average value of (2);Nrepresenting a set of neighbor users of the target user; />For usersuAnd a uservOverall similarity of (c); />And->Respectively represent neighboring usersvFor articlesiIs a comprehensive score and user of (2)vIs a comprehensive score average value of (2);
based on the predicted comprehensive scores of the user on the items, arranging the items, and selecting top-N items as the user to form a recommendation result.
(4) And establishing a user evaluation feedback model based on the eye movement signals, acquiring the evaluation of the user on the recommendation result, and forming feedback information to correct the user preference.
At present, the performance of a recommendation algorithm is generally measured through indexes such as recall rate and accuracy rate, the indexes are obtained through offline experiment calculation, the overall performance of the recommendation algorithm can be evaluated, but the satisfaction degree of a user on a recommendation result cannot be judged, and personalized evaluation of the user on a recommendation system cannot be obtained. In addition, the current recommendation algorithm generally only uses implicit information to perform feedforward calculation and lacks a feedback mechanism, so that personalized evaluation information of a user can be used as feedback information to update interaction information of the user and a project, modify user preference and further continuously optimize the recommendation algorithm.
When a user browses TOP-N items of a recommendation result, a user evaluation feedback model is established by using a user real-time eye movement signal and an SVR model, personalized evaluation of each TOP-N item by the user is obtained, the evaluation result is used as feedback information, interaction information between the user and the items is updated, user preference is corrected, and a recommendation algorithm is optimized, so that the recommendation result can meet real requirements of the user.
Next, the technical scheme disclosed in the embodiment and the achieved beneficial effects are further known based on specific experiments.
Performing an experiment of neighbor user optimization based on user multidimensional preference clustering based on a common data set (TA-data), wherein the experiment comprises the following steps: user clustering experiments and PSO user preference weight optimizing experiments. The TA-data set comprises 10273 data records of 539 hotels about sleeping, services, rooms, positions, cleanliness and prices of the 1595 users in total; a scoring interval between 1 and 5; the data sparsity was 98.81%.
(1) User clustering experiment
Based on the relation between the user multidimensional scores and the total scores, the multidimensional preferences of the users are mined and analyzed, and clustering is carried out based on the user preferences, and the effect of the K-means clustering algorithm is sensitive to the clustering quantity, so that the proper clustering quantity is selected. The comparison of the cluster profile coefficients is shown in fig. 6 for different numbers of clusters.
As can be seen from the experimental results, the user profile coefficient is larger when the number of clusters is 9, 16 and 19, and the average profile coefficient is larger when the number of clusters is 9 according to the elbow method and experimental analysis, namelySC=0.35, the clustering effect is better. After clustering, user preferences of the same category may be considered to have similarity.
(2) PSO user preference weight optimizing experiment
After the user clustering is completed, the similarity of the users in a plurality of dimensions under each category needs to be calculated, and then PSO is utilized to optimize the fusion weight of each similarity, so that more accurate overall similarity is obtained.
As shown in fig. 7, which is a multi-dimensional similarity fusion weight optimizing result diagram of one type of users, it can be found that when the iteration number reaches about 51, the fitness function value is not changed any more, the corresponding fitness value, that is, MAE is 0.768, and pso completes optimizing.
Further, in addition to the technical solution disclosed in this embodiment, the user attention may be mined based on other factor information, for example: information such as brain electricity, electrocardio, skin movement, expression and the like; the user attention can be analyzed based on eye movement signals or other human factor information, and the user similarity is calculated according to the user attention, so that the neighbor user set is optimized; the relationship between eye movement signals or other human factor information and the attention of the user can be constructed through other regression models; the user multidimensional information and the user attention can be subjected to fusion analysis through other aggregation models or algorithms; according to eye movement signals or other human factor information, a user evaluation feedback mechanism can be established to improve and optimize other processes or modules in the recommendation algorithm, such as: score prediction, similarity calculation, matrix decomposition model, and the like.
In the present specification, each embodiment is described in a progressive manner, and each embodiment is mainly described in a different point from other embodiments, and identical and similar parts between the embodiments are all enough to refer to each other.
The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims (6)
1. The multidimensional recommendation method combining the human factor information is characterized by comprising the following steps of:
determining neighbor users based on the multi-dimensional similarity fused by the preference weights of the users;
constructing a user attention model based on the eye movement signal, and acquiring the attention of a user when browsing the project;
based on the depth neural network, merging the multidimensional scores and the attention degree, and generating a recommendation result;
constructing a user evaluation feedback model based on the eye movement signals, acquiring the evaluation of the user on the recommendation result, and forming feedback information to correct the user preference;
the method for acquiring the attention degree of the user when browsing the items is as follows: analyzing interaction behaviors and eye movement signals when a user browses items by utilizing the eye movement signals, and establishing a nonlinear relation between the eye movement signals and the attention of the user; the method specifically comprises the following steps:
let s= { (x) i ,y i ) Is input sample data, where y i E R represents user attention, x i ∈R N As eye movement data samples, four eye movement index features including gazing time length, gazing point number, pupil diameter and eye jump number are adopted as sample data through correlation regression analysis;
the expression of the regression function is:
wherein:is a nonlinear mapping function and is used for mapping training data to a high-dimensional linear space; omega and b are model parameters;
introducing a relaxation variable ζ i Andconverting the nonlinear support vector regression problem into a quadratic programming problem:
wherein: zeta type toy i ，i (i=1, 2, …, n) is the number of samples, C' is a penalty factor, epsilon is a insensitivity loss coefficient;
a Lagrange multiplier is introduced, and parameters in the quadratic programming problem are solved, so that the attention of a user to the project is obtained;
the method comprises the steps of obtaining the evaluation of a user on a recommendation result, forming feedback information to correct user preference, and specifically comprising the following steps:
constructing a user evaluation feedback model based on the eye movement signals, tracking and analyzing the eye movement signals when the user browses the recommendation result in real time, and acquiring personalized evaluation of the user about the recommendation result;
and taking the evaluation result as feedback information, correcting user preference, updating interaction information of the user and the project, and optimizing a recommendation algorithm.
2. The multi-dimensional recommendation method combining human factor information according to claim 1, wherein determining the neighboring user specifically comprises the steps of:
based on the multidimensional scoring information of the users, carrying out user classification by adopting a clustering algorithm;
and calculating the similarity of each dimension of various users, carrying out multi-dimension similarity weight optimization of the users by adopting a particle swarm optimization algorithm, and obtaining the total similarity of the users by weighting and fusion to obtain a neighbor user set.
3. The multi-dimensional recommendation method combining human factor information according to claim 2, wherein the specific process of user classification is:
based on the Euclidean distance between the multidimensional score and the total score of the user, obtaining a preference value of each user:
wherein: u is the target user, r 0 Scoring the population of items for the target user, r k Scoring a kth dimension of the item for the target user; i (u) is a set of items scored for user u; p is p uk Preference value for user u for kth dimension; d, d u (r 0 ,r k ) The Euclidean distance between the multidimensional scores and the total scores of the users is calculated;
calculating preference values of all users for each dimension, and carrying out normalization processing to obtain preference characteristics of the users;
based on preference characteristics of users, the users are clustered by adopting a K-means algorithm.
4. The multidimensional recommendation method combining human factor information according to claim 2, wherein the specific process of obtaining the neighbor set of users is as follows:
and calculating the similarity of each dimension between users in the same category by adopting the Pelson coefficient:
wherein: sim (Sim) c (u, v) represents the similarity of user u and user v in dimension c; r is R c (u, i) and R c (v, i) represent user u and user v scoring item i in dimension c, respectively; i u,v A set of items representing a common score for user u and user v;andrepresenting the average score of user u and user v in the set of items with respect to dimension c, respectively;
and carrying out similarity fusion according to the preference characteristics of the user, wherein the formula of the similarity fusion of the user is as follows:
wherein: sim (u, v) is the overall similarity of user u and user v, sim i (u, v) and w i The similarity of each dimension of the user u and the user v and the corresponding weight are respectively represented;
optimizing the similarity fusion weight of each dimension by adopting a particle swarm optimization algorithm, taking the minimum value of the average absolute error between the predicted score and the actual score of the user as an optimized objective function, wherein the fitness function is as follows:
wherein: p (u, i) andthe actual score and the predicted score of the user u and the user v on the item i are respectively represented, and N is the number of the predicted items.
5. The multi-dimensional recommendation method combining human factor information according to claim 1, wherein generating the recommendation result comprises the following steps:
processing and fusing the multi-dimensional scoring information of the users in each category and the attention degree of the users by using the deep neural network to obtain the comprehensive scoring of the users on the items;
in the neighbor user set, predicting the comprehensive score of the target user on the non-contact item by a collaborative filtering algorithm, wherein the calculation formula is as follows:
wherein: p (P) ui Representing a predictive score of the object i by the target user u;representing the comprehensive score mean value of the target user u; n represents the neighbor user set of the target user; sim (u, v) is the overall similarity of user u and user v; r is R vi And->Respectively representing the comprehensive score of the neighbor user v on the object i and the average value of the comprehensive scores of the users v;
and sequencing all the items based on the predicted comprehensive scores of the items by the user, and selecting top-N items as recommendation results formed by the user.
6. The multidimensional recommendation method combining human factor information according to claim 1, wherein the input layer of the deep neural network is a multidimensional score and a degree of attention of a user, the hidden layer is a 2-layer, and the output of the output layer is a comprehensive score of the user.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311026200.XA CN116738072B (en) | 2023-08-15 | 2023-08-15 | Multidimensional recommendation method combining human factor information |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311026200.XA CN116738072B (en) | 2023-08-15 | 2023-08-15 | Multidimensional recommendation method combining human factor information |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116738072A CN116738072A (en) | 2023-09-12 |
| CN116738072B true CN116738072B (en) | 2023-11-14 |
Family
ID=87911892
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311026200.XA Active CN116738072B (en) | 2023-08-15 | 2023-08-15 | Multidimensional recommendation method combining human factor information |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116738072B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117171708B (en) * | 2023-11-01 | 2024-02-23 | 小舟科技有限公司 | Multimode fusion method, system, equipment and medium in hybrid BCI system |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106897363A (en) * | 2017-01-11 | 2017-06-27 | 同济大学 | The text for moving tracking based on eye recommends method |
| CN112699295A (en) * | 2019-10-23 | 2021-04-23 | 腾讯科技（深圳）有限公司 | Webpage content recommendation method and device and computer readable storage medium |
| CN113378047A (en) * | 2021-06-10 | 2021-09-10 | 武汉大学 | Multi-aspect enhancement-based graph neural network recommendation method |
| WO2022011448A1 (en) * | 2020-07-16 | 2022-01-20 | Lululemon Athletica Canada Inc. | Method and system for an interface for personalization or recommendation of products |
| CN114399367A (en) * | 2022-01-24 | 2022-04-26 | 平安科技（深圳）有限公司 | Insurance product recommendation method, device, equipment and storage medium |
| CN114547480A (en) * | 2022-02-22 | 2022-05-27 | 创络(上海)数据科技有限公司 | Deep learning recommendation method and system based on multi-platform fusion |
| CN115907869A (en) * | 2023-03-08 | 2023-04-04 | 江西财经大学 | Attention degree-based elevator advertisement recommendation method and system |
-
2023
- 2023-08-15 CN CN202311026200.XA patent/CN116738072B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106897363A (en) * | 2017-01-11 | 2017-06-27 | 同济大学 | The text for moving tracking based on eye recommends method |
| CN112699295A (en) * | 2019-10-23 | 2021-04-23 | 腾讯科技（深圳）有限公司 | Webpage content recommendation method and device and computer readable storage medium |
| WO2022011448A1 (en) * | 2020-07-16 | 2022-01-20 | Lululemon Athletica Canada Inc. | Method and system for an interface for personalization or recommendation of products |
| CN113378047A (en) * | 2021-06-10 | 2021-09-10 | 武汉大学 | Multi-aspect enhancement-based graph neural network recommendation method |
| CN114399367A (en) * | 2022-01-24 | 2022-04-26 | 平安科技（深圳）有限公司 | Insurance product recommendation method, device, equipment and storage medium |
| CN114547480A (en) * | 2022-02-22 | 2022-05-27 | 创络(上海)数据科技有限公司 | Deep learning recommendation method and system based on multi-platform fusion |
| CN115907869A (en) * | 2023-03-08 | 2023-04-04 | 江西财经大学 | Attention degree-based elevator advertisement recommendation method and system |
Non-Patent Citations (1)
| Title |
|---|
| 基于ResNet神经网络的情报用户隐式反馈关注度预测;吴俊杰 等;《空天预警研究学报》;第第36卷卷(第第3期期);201-205 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116738072A (en) | 2023-09-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112598462B (en) | Personalized recommendation method and system based on collaborative filtering and deep learning | |
| Zhou et al. | Atrank: An attention-based user behavior modeling framework for recommendation | |
| CN107423442B (en) | Application recommendation method and system based on user portrait behavior analysis, storage medium and computer equipment | |
| He et al. | Ups and downs: Modeling the visual evolution of fashion trends with one-class collaborative filtering | |
| CN116738072B (en) | Multidimensional recommendation method combining human factor information | |
| Jiao et al. | A novel learning rate function and its application on the SVD++ recommendation algorithm | |
| CN110866145A (en) | Co-preference assisted deep single-class collaborative filtering recommendation method | |
| CN112529638B (en) | Service demand dynamic prediction method and system based on user classification and deep learning | |
| Hao et al. | Annular-graph attention model for personalized sequential recommendation | |
| CN108615177A (en) | Electric terminal personalized recommendation method based on weighting extraction interest-degree | |
| CN113918833A (en) | Product recommendation method realized through graph convolution collaborative filtering of social network relationship | |
| Chen et al. | DeepGly: A deep learning framework with recurrent and convolutional neural networks to identify protein glycation sites from imbalanced data | |
| CN113051468B (en) | Movie recommendation method and system based on knowledge graph and reinforcement learning | |
| Qin et al. | Towards a personalized movie recommendation system: A deep learning approach | |
| Guimarães et al. | Guard: A genetic unified approach for recommendation | |
| Xu et al. | Exploiting interactions of review text, hidden user communities and item groups, and time for collaborative filtering | |
| Rawat et al. | An embedding-based deep learning approach for movie recommendation | |
| KR20220125422A (en) | Method and device of celebrity identification based on image classification | |
| Xue et al. | An Explicit Feedback Recommendation Algorithm Based on Subjective and Objective Evaluation Transformation Model | |
| Liu et al. | Research on Personalized Recommendation Model and algorithm in deep learning mode | |
| Almu et al. | Incorporating preference Changes through users’ input in collaborative filtering movie recommender system | |
| Wang et al. | An improved constrained Bayesian probabilistic matrix factorization algorithm | |
| Sridhar et al. | Extending Deep Neural Categorisation Models for Recommendations by Applying Gradient Based Learning | |
| Feng | To become fashionable: A brief review of outfit compatibility | |
| CN114117251B (en) | Intelligent context-Bo-down fusion multi-factor matrix decomposition personalized recommendation method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |